<div itemprop="articleBody" data-v-0e838cd6="" style="font-weight: 400;"><p><strong>点赞 + 关注 + 收藏 = 学会了</strong></p>
<h1 data-id="heading-0">本文简介</h1>
<p>千盼万盼，终于等到 Stability AI 开源了 Stable Diffusion 3 Medium 模型了。这个号称比 MJ 更牛的模型可以在民用电脑跑起来。本文简单介绍怎么才能用上这个模型~</p>
<p>先来看看官方给出的样张。</p>
<p><img src="https://p6-juejin.byteimg.com/tos-cn-i-k3u1fbpfcp/5790bfa9193b431794ba4a9a9f806b05~tplv-k3u1fbpfcp-jj-mark:3024:0:0:0:q75.awebp#?w=3072&h=2820&s=7404768&e=jpg&b=dfcebc" alt="01.jpg" loading="lazy"></p>
<p>闲话不多说，先看看怎么用起来。</p>
<p>简简单单只需几步：</p>
<ol>
<li>下载 ComfyUI</li>
<li>下载 stable-diffusion-3-medium 模型</li>
<li>将 stable-diffusion-3-medium 放在 ComfyUI 的模型目录里</li>
<li>运行 ComfyUI 画图</li>
</ol>
<h1 data-id="heading-1">动手操作</h1>
<h2 data-id="heading-2">1、下载 ComfyUI</h2>
<p>ComfyUI下载地址：<a href="https://link.juejin.cn/?target=https%3A%2F%2Fgithub.com%2Fcomfyanonymous%2FComfyUI%2Freleases" target="_blank" title="https://github.com/comfyanonymous/ComfyUI/releases" ref="nofollow noopener noreferrer">github.com/comfyanonym…</a></p>
<p><img src="https://p9-juejin.byteimg.com/tos-cn-i-k3u1fbpfcp/22b55944fae3400e8d42b079eb73e8d8~tplv-k3u1fbpfcp-jj-mark:3024:0:0:0:q75.awebp#?w=1469&h=729&s=250688&e=jpg&b=fffefe" alt="02.png" loading="lazy"></p>
<p>点击“Download Link with stable pytorch 2.3 cu121”下载。</p>
<p>下载完会得到一个“ComfyUI_windows_portable_nvidia_cu121_or_cpu.7z”文件，把它解压出来。</p>
<h2 data-id="heading-3">2、下载 stable-diffusion-3-medium 模型</h2>
<p>第2步是下载 stable-diffusion-3-medium 模型。</p>
<p>下载地址：<a href="https://link.juejin.cn/?target=https%3A%2F%2Fhuggingface.co%2Fstabilityai%2Fstable-diffusion-3-medium%2Ftree%2Fmain" target="_blank" title="https://huggingface.co/stabilityai/stable-diffusion-3-medium/tree/main" ref="nofollow noopener noreferrer">huggingface.co/stabilityai…</a></p>
<p><img src="https://p3-juejin.byteimg.com/tos-cn-i-k3u1fbpfcp/5429b568b5594e24b2c3a1f7e70765c5~tplv-k3u1fbpfcp-jj-mark:3024:0:0:0:q75.awebp#?w=1920&h=912&s=410901&e=jpg&b=ffffff" alt="03.png" loading="lazy"></p>
<p>红框框住的几个模型都可以用，但它们的体积不一样。简单来说，体积越大的模型就越强，但也更吃性能。根据你自己的需求选模型。</p>
<h2 data-id="heading-4">3、把模型放在 ComfyUI 的模型目录里</h2>
<p>将你下载好的模型放在 <code>ComfyUI_windows_portable\ComfyUI\models\checkpoints</code> 这个目录里。</p>
<p><img src="https://p6-juejin.byteimg.com/tos-cn-i-k3u1fbpfcp/2183ef44c90640a3b3b26230f6c2dcc7~tplv-k3u1fbpfcp-jj-mark:3024:0:0:0:q75.awebp#?w=1111&h=369&s=123707&e=jpg&b=fefdfd" alt="04.png" loading="lazy"></p>
<h2 data-id="heading-5">4、运行 ComfyUI 画图</h2>
<p>回到 ComfyUI 的根目录。</p>
<p><img src="https://p9-juejin.byteimg.com/tos-cn-i-k3u1fbpfcp/69e70ec0e55d4317b85ff9a5204473cb~tplv-k3u1fbpfcp-jj-mark:3024:0:0:0:q75.awebp#?w=326&h=325&s=28796&e=jpg&b=fffefe" alt="05.png" loading="lazy"></p>
<p>如果你没有N卡，那就双击运行 <code>run_cpu.bat</code> 文件。</p>
<p>如果你有N卡就运行 <code>run_nvidia_gpu.bat</code> 文件。</p>
<p>运行后会弹出一个终端，等它运行几秒后，浏览器会自动打开 <code>http://127.0.0.1:8188/</code>。</p>
<p>界面长这样。</p>
<p><img src="https://p9-juejin.byteimg.com/tos-cn-i-k3u1fbpfcp/14546df727854aa6b0b3dd421020822f~tplv-k3u1fbpfcp-jj-mark:3024:0:0:0:q75.awebp#?w=1920&h=912&s=367941&e=jpg&b=252525" alt="06.png" loading="lazy"></p>
<p>我试了一下 Stability 前段时间发布的样张里的提示词，用的是 <code>sd3_medium_incl_clips_t5xxlfp8.safetensors</code> 这个模型。</p>
<pre><code lang="css"><span data-line-num="1">cinematic photo of <span>a</span> red apple on <span>a</span> <span>table</span> in <span>a</span> classroom, on the blackboard are the words “go big or go home” written in chalk.</span>
</code></pre>
<p><img src="https://p1-juejin.byteimg.com/tos-cn-i-k3u1fbpfcp/5d6f2d1ae2824f309be0e04cfb3abec3~tplv-k3u1fbpfcp-jj-mark:3024:0:0:0:q75.awebp#?w=1536&h=1024&s=1577121&e=png&b=0a3316" alt="07.png" loading="lazy" class="reader-image-tiny"></p>
<p>对文字的理解感觉没有官方吹的那么厉害，但相比起之前的 V1.5 模型来说已经有非常明显的进步了。</p>